Speech Recognition Software

[ Home ] [ Site Map ] [ Site Search ] [ Back to last page ]

On this page I will tell you about my experiences with speech recognition software.  This is a fairly new area to me, but since I am such a lazy/poor typist, I decided that it was about time I did something about it...hehe...and since I'm a geek, rather than using a typing program to become more proficient, I decided to get speech recognition software.

I have fooled around with various speech recognition engines over the last few years, but I've never really been very happy with them.  Thanks to a few PGCN newsletter subscribers, I finally decided to go full speed ahead and really give it a try.

From everything that I have read, as well as from everyone I've talked to, Dragon Systems "Dragon NaturallySpeaking" is the only software to use.  So I went out and bought Dragon NaturallySpeaking version 5 preferred edition (I don't really know what the preferred edition actually does, but since it was the top-of-the-line, that's what I got...hehe).

As soon as I got home after buying the software, I decided to try it out on my brand-new Dell laptop.  Since the software came with this really nifty headset/microphone I thought I was all set...hehe...so I installed the software, plugged in the microphone, fired up the software, and started a five-minute training session.  Boy did my eyes get opened!  The software didn't seem to be able to understand half the words I spoke. Anytime I stopped to take a breath it would start typing all sorts of strange words and letter combinations all by itself.  After fighting with this for about a hour I gave up in complete disgust!.

So what did I do next?  I went to the Dragon Systems website, and started reading.  The first thing I found out was that the sound card on my brand new Dell laptop was not on the approved hardware list.  Then, as I continued reading I discovered that the quality of the microphone was critical to the success of any form of speech recognition software.

Now think about this for second...a headset/microphone was packaged with my software, yet according to their website, one of the biggest issues with low accuracy is caused by a low quality microphone.  So I asked myself, "Self, why would they package a cheap microphone with their product knowing full well it wouldn't work worth a hoot?"  Sometimes I think those marketing boys really don't have any idea what they're doing.

My first concern was the fact that the sound card on my Dell laptop was not on their approved hardware list.  Since I just purchased the laptop recently I thought "game over man"!  One of the PGCN subscribers specifically mentioned a "digital microphone".  So I started to do a search on the Web using my favorite search engine, "Google", and after about 10 minutes I landed on this really awesome site called "SpeechControl".

As far as I'm concerned, if you have any interest in speech recognition software, this is the only website you should go to when it comes to microphones and accessories for that software.  Every piece of equipment that they sell has been thoroughly tested by them.  In my view, they've taken all the guesswork out of it for me!

After looking around on their website for little bit, I found that I really didn't have enough knowledge to make an intelligent purchasing decision.  So I went to the contact page and found I could send an e-mail to them and they would call me back.

Now let me tell you, this really surprised me!  It was around 8 o'clock at night when I sent the e-mail and Martin Markoe called me back within about half an hour.  I explained that I was using a laptop and would be working in a noisy environment.  I also explained that I had some concerns since I have a slight speech impediment.  Martin came across very knowledgeable, he knew his products well, he knew their prices, and he knew that they had the products in stock.

As for my concerns regarding my speech impediment, he explained that while I might need to do a little bit of additional training, the software would learn my speech patterns over time and this would not be an issue.  At the end of the call I told Martin that I would think about it and get back to him and thanked him for his time.

So I thought about it, and talked to the Boss (Amy, my wife).  The next day I sent Martin another e-mail asking him to call me which he did in about 15 minutes...wow...and told him what parts I had landed on.  I asked him if they were in stock, he said they were, and I explained my concerns about having to wait for them to be shipped.  He indicated that they should ship the same day ordered, and I should receive them in approximately three days.

After again thanking him for his time I immediately placed the order on the website.  About three hours later I received a confirmation e-mail, and about an hour after that I received another e-mail stating that the order had been shipped.

So three days later I ended up with my brand-new headset/microphone and a really cool USB sound card (the sound card requires no software to be installed in either Windows 98 SE or Windows ME.  Just plug it in and it worked).  I fired up the speech recognition software, deleted the current user, created a new user, and did the five-minute training session.  At that point I had about an 85 percent accuracy and most importantly, no weird characters or phrases ending up on the screen!

To my thinking an 85 percent accuracy level isn't anything to get excited about.  So at that point I decided it was time to start reading the manual...hehe...and I learned some really interesting stuff.  Besides doing advanced training, you can also have the program read documents that you have all ready written to find new words to add to the vocabulary as well as to learn your actual writing style.  So I had the program read about 15 pages from PuterGeek .Com as well as about 20 e-mails that I had written.  After that, I then did two advanced training sessions which lasted approximately 40 minutes and immediately started working on this web page.

Yup, believe it or not, this entire web page was written with Dragon NaturallySpeaking (version 5, preferred edition)!  What I have written above would normally have taken me at least 2 to 2 1/2 hours to get done.  But thanks to this neat software, I've only spent about 30 minutes so far.  True, I have had to make some corrections.  Some of the words it wrote really made no sense whatsoever, but I would truthfully say that my accuracy is currently around 90 percent or so.  Also, it does a much better job of spelling than I ever could as well (grin)!

I have been told that it is quite reasonable to expect the software to get somewhere around a 98 percent accuracy with practice and correction.  Yes, it will take a small investment of my time, but considering how poor a typist I am, I feel it will be time very well spent!

This page is being written in FrontPage 2000, and I have found that some of the NaturallySpeaking features are not available while working inside FrontPage.  For instance, whenever I start a new paragraph the first word is never capitalized.  If I need to correct a mistyped word, I have to do it manually.  I have yet to use this in Outlook Express but I have high hopes for it.

Here is what I spent to get all this up to speed.

Obviously, if you took a little bit of time looking for better prices, I'm sure you could do better.  But I simply don't have the time.  Plus the guys at Speech Controlled Computer Systems (SCCS) did all the work for me!  They figured out which headset/microphone I should buy as well as resolving my sound card issue.  In fact, this whole web page has been written on my laptop, in the truck while idling at 900 RPM (it is 105 degrees in Pecos, TX right now), and the background noise has not been an issue at all tonight.

A couple of notes on performance...while I believe that a 700 MHz Celeron processor is fast enough, I don't think that 128 megs of ram is quite enough.  With FrontPage 2000 and NaturallySpeaking open (WinME), System Monitor shows 8 megs of ram in the disk cache, 16 megs of ram in swap file, and 10 megs of ram free.  As you know, anytime you hit the swap file you slow down the 'puter.  Normally, while connected to the internet and working on the website, I tend to have three to four browser windows open as well as my FTP program.  In this case, I would be hitting the swap file quite a bit.  In my opinion, the average power user should plan to have at least 256 megs of ram when using this software.  As with anything else 'puter related, more CPU and more memory is always better.

In closing, I believe that Dragon NaturallySpeaking, combined with a good quality headset microphone, will do the job exactly as expected.  Sure, I'll have to spend little bit of time and effort in training, but it'll be worth it to me.  Like most people, I have yet to read the manual completely nor have I bothered to learn all of the commands yet.  As I become more proficient with the software, and it becomes more used to my writing style, I fully expect to be able to create web pages and answer my e-mail in substantially less time and with far greater ease.

I strongly suggest that you take a look at the SpeechControl website!  I found Martin to be very friendly, as well as willing to spend some time and answer all of my questions.  I have no problem whatsoever with recommending him to all of you.

[ Home ] [ Site Map ] [ Site Search ] [ Back to last page ]

Last Revised: 05/26/2001